Therapeutic drug monitoring in plasma and CSF

Therapeutic drug monitoring usually measures blood and stops there. A validated UPLC-MS/MS method run on matched plasma and cerebrospinal fluid from 20 patients found no link between plasma concentration and tumor response, but a strong link between cerebrospinal fluid exposure and intracranial response. The matrix choice was the whole result.

Why therapeutic drug monitoring needs the right matrix

Most monitoring programs measure plasma or serum because it is easy to obtain and reference ranges exist. That convenience becomes a problem when the target site sits behind a barrier, because plasma concentration is then a proxy for something it does not actually measure.

The blood-brain barrier is the clearest case. A drug can sit at a perfectly adequate plasma trough and still deliver a small fraction of that concentration to the central nervous system. Penetration depends on molecular weight, lipid solubility, efflux transporter activity and barrier integrity, so two patients on the same dose can differ by more than ten-fold in target-site exposure.

Cerebrospinal fluid sampling is invasive and low volume, which is why it is rarely done. But the penetration rate, calculated as cerebrospinal fluid concentration divided by plasma concentration and multiplied by 100, is the parameter that explains response when plasma cannot.

QuestionPlasma answerCerebrospinal fluid answer
Was the dose absorbedYesYes
Is systemic exposure adequateYesOnly indirectly
Did the drug reach the target siteNoYes
Does exposure track with responseOften notIn this cohort yes
Sample volume neededLargeSmall and precious

What the validated method had to prove

The method was built to quantify two third-generation EGFR inhibitors, osimertinib and almonertinib, in a single run in two matrices. Running both drugs and both matrices in one method is the point, because it allows a direct head-to-head comparison of central nervous system penetration without cross-assay calibration differences.

Any therapeutic drug monitoring method has to prove this much before a single patient sample is run. Samples came from 20 patients with non-small cell lung cancer and brain metastases at steady-state trough, 7 on osimertinib and 13 on almonertinib. Response was assessed by RECIST criteria on lung CT and brain MRI, and penetration was calculated for every matched pair.

Validation parameterAcceptance standard usedResult
SelectivityInterference below 20 percent of LLOQ and 5 percent of internal standardMet in both matrices
LinearityCorrelation coefficient across the calibration range0.994 to 0.998
Lower limit of quantitationDefined as the lowest calibrator1 ng/mL in both matrices
PrecisionRSD at low, medium and high QC levels15 percent or below
AccuracyRelative error at QC levelsWithin plus or minus 10 percent
Matrix effectResponse ratio in six blank matrix lots86.98 to 104.23 percent
RecoveryExtraction recovery across QC levels101.15 to 106.32 percent

Selectivity calibration and the lower limit

Retention times were consistent between matrices, which matters when one method runs both. In plasma, almonertinib, the internal standard and osimertinib eluted at 1.351, 2.198 and 1.512 minutes. Corresponding times in cerebrospinal fluid were 1.350, 2.197 and 1.517 minutes. A shift of a few thousandths of a minute between matrices is what you want to see.

Selectivity was shown by comparing interfering response at the analyte retention time against the response at the lower limit of quantitation. Interference stayed below 20 percent of the LLOQ response and below 5 percent of the internal standard response, so endogenous matrix components did not compromise quantitation.

DrugMatrixCalibration rangeCorrelation coefficient
OsimertinibPlasma1 to 1,000 ng/mL0.995
OsimertinibCerebrospinal fluid1 to 100 ng/mL0.997
AlmonertinibPlasma1 to 1,000 ng/mL0.994
AlmonertinibCerebrospinal fluid1 to 100 ng/mL0.998

Note that the cerebrospinal fluid range is ten times narrower than the plasma range. That is the right design, because cerebrospinal fluid concentrations here run one to two orders of magnitude below plasma, and forcing a shared range would either waste plasma capacity or push the cerebrospinal fluid samples below the lowest calibrator.

Precision accuracy matrix effect and recovery

Quality control samples sat at 3, 100 and 800 ng/mL in plasma and 3, 10 and 80 ng/mL in cerebrospinal fluid. Intra-batch and inter-batch precision came in at 15 percent RSD or below, and accuracy stayed within plus or minus 10 percent relative error, which meets the pharmacopoeial criteria for bioanalytical method validation.

Matrix effect was calculated from six blank lots of each matrix at low and high concentration. Values ran from 86.98 to 104.23 percent, inside the accepted window for quantitative bioanalytical work, so ion suppression or enhancement did not distort the measured concentrations.

Extraction recovery ranged from 101.15 to 106.32 percent across levels, with mean recovery of 104.1 percent in plasma and 102.9 percent in cerebrospinal fluid. Recovery that high and that consistent across two matrices is the sign of a sample preparation step that is not the weak link.

Stability under real storage conditions

Stability testing covered the conditions a clinical sample actually sees. Low and high concentration samples in both matrices were held for 6 hours at 25 degrees Celsius, for 30 days at minus 80 degrees Celsius, and through four freeze-thaw cycles at minus 80 degrees Celsius.

Across all of those conditions, precision stayed at 15 percent RSD or below and relative deviation stayed within plus or minus 15 percent of nominal. The long-term and freeze-thaw arms matter most in practice, because retrospective cerebrospinal fluid studies depend entirely on banked samples that have already been through several cycles.

What the paired plasma and CSF samples showed

Inter-individual variability was large in every dimension: plasma concentration, cerebrospinal fluid concentration and penetration rate all spread widely. For osimertinib, mean plasma concentration was 179.46 ng/mL with a range of 45.30 to 395.87, and mean cerebrospinal fluid concentration was 6.35 ng/mL.

Penetration ratios for osimertinib ranged from 1.21 to 6.29 percent with a mean of 3.23 percent. For almonertinib the range was far wider, 0.34 to 30.86 percent, with a mean of 5.92 percent and a mean cerebrospinal fluid concentration of 8.39 ng/mL.

DrugMean plasmaMean CSFMean penetrationPenetration range
Osimertinib179.46 ng/mL6.35 ng/mL3.23 percent1.21 to 6.29 percent
Almonertinibnot reported8.39 ng/mL5.92 percent0.34 to 30.86 percent

The counterintuitive part: plasma exposure did not predict response

Steady-state plasma concentration showed no significant correlation with lung tumor response for either drug. That part is not surprising, since a plasma trough failing to predict a lung lesion is a common enough finding.

The informative result is the split between the two compartments. For brain MRI response, patients achieving partial response had significantly higher cerebrospinal fluid concentrations, with P below 0.001, and higher penetration ratios, with P below 0.01, than patients with stable or progressive disease. No significant difference appeared between stable and progressive disease groups.

The single clearest case was a patient who achieved partial response in the lung while intracranial disease progressed, with a cerebrospinal fluid concentration of 1.1 ng/mL and a penetration rate of 0.44 percent. Adequate systemic exposure and near-absent central nervous system exposure in one person is the argument for paired sampling in a single number.

Caution is required on the statistics. The response groups were small, 3 partial response, 7 stable disease and 3 progressive disease, and no statistically significant differences were detected among response groups for osimertinib alone. The correlation is real in this dataset and fragile in its sample size.

What therapeutic drug monitoring looks like for acyclovir valganciclovir and mycophenolate mofetil

The same design question applies well outside oncology. Acyclovir and its prodrug valacyclovir are hydrophilic nucleoside analogues with well-characterized renal handling, and monitoring matters most when clearance is impaired rather than when absorption is uncertain. Valganciclovir, the prodrug of ganciclovir, is monitored in transplant recipients because both under-exposure and over-exposure carry real consequences, and the marrow toxicity of ganciclovir sets an upper bound as well as a lower one.

Mycophenolate mofetil is the clearest candidate of the three. Mycophenolic acid exposure varies several-fold between patients on the same dose, protein binding shifts with renal function and albumin, and the enterohepatic recirculation contribution makes a single trough a poor estimator of total exposure. Limited sampling strategies and area under the curve estimation exist for exactly that reason.

The transferable lesson is not about the specific drug. It is that the monitoring matrix has to match the site of action, and that a validated method has to be proven in that matrix rather than assumed to carry over from plasma.

DrugWhat exposure varies withWhat the monitoring question really is
AcyclovirRenal clearanceIs clearance impaired enough to require dose adjustment
ValganciclovirRenal function and marrow toleranceIs exposure inside the window between efficacy and toxicity
Mycophenolate mofetilProtein binding, recirculation, co-medicationDoes total exposure match the intended immunosuppressive load

Where the evidence stops

Twenty patients is small, and the response subgroups have 3, 7 and 3 patients. Several comparisons did not reach significance, including every plasma-based comparison and the cerebrospinal fluid comparisons within the osimertinib group alone.

Cerebrospinal fluid concentration is a surrogate for brain tissue concentration, not a direct measure of it. Lumbar sampling introduces its own variability, and a single trough says nothing about the concentration-time profile at the target site.

The penetration rate is a ratio of two measurements, so it inherits the error of both. That is acceptable for a research question and less acceptable for an individual dosing decision, where two noisy numbers divided by each other can mislead.

Therapeutic drug monitoring in cerebrospinal fluid is not a routine service today, and making it one would require a validated target range first. Finally, none of this is a dosing rule. The study shows association between cerebrospinal fluid exposure and imaging response, and it was not designed to show that adjusting the dose to hit a cerebrospinal fluid target improves outcomes.

Bottom line

A validated UPLC-MS/MS method quantified two EGFR inhibitors in matched plasma and cerebrospinal fluid with precision at or below 15 percent RSD, accuracy within plus or minus 10 percent, matrix effect of 86.98 to 104.23 percent and recovery above 100 percent in both matrices. Plasma concentration did not track with response; cerebrospinal fluid concentration and penetration rate did, with P below 0.001 and P below 0.01 respectively. Therapeutic drug monitoring answers the question you ask it, and the matrix decides which question that is.

FAQ

What is therapeutic drug monitoring

Therapeutic drug monitoring measures drug concentration in a biological fluid to guide dosing rather than relying on dose alone. It is used when exposure varies widely between patients on the same regimen, or when the gap between efficacy and toxicity is narrow. The fluid chosen determines what the number actually means.

Why measure cerebrospinal fluid instead of plasma

Plasma is a proxy, and behind a barrier it can be a poor one. Penetration depends on molecular weight, lipid solubility, efflux transporter activity and barrier integrity, so two patients at the same plasma concentration can differ by more than ten-fold at the target site. Cerebrospinal fluid measures the compartment that matters.

How is the blood-brain barrier penetration rate calculated

Penetration rate is cerebrospinal fluid concentration divided by plasma concentration, multiplied by 100, expressed as a percent. In this cohort osimertinib ranged from 1.21 to 6.29 percent with a mean of 3.23 percent, and almonertinib from 0.34 to 30.86 percent with a mean of 5.92 percent. Each figure is a ratio of two measurements, so it carries the error of both.

What precision and accuracy did the method achieve

Intra-batch and inter-batch precision was 15 percent RSD or below at low, medium and high quality control levels, and accuracy stayed within plus or minus 10 percent relative error. Those figures meet the pharmacopoeial criteria for bioanalytical method validation in both plasma and cerebrospinal fluid.

What was the lower limit of quantitation

The lower limit of quantitation was 1 ng/mL in both plasma and cerebrospinal fluid for both drugs. Calibration ran from 1 to 1,000 ng/mL in plasma and from 1 to 100 ng/mL in cerebrospinal fluid, a ten-fold narrower range that matches the much lower concentrations seen in cerebrospinal fluid. A shared range would either waste plasma capacity or push samples below the lowest calibrator.

How was selectivity demonstrated

Interfering response at the analyte retention time stayed below 20 percent of the response at the lower limit of quantitation and below 5 percent of the internal standard response. Retention times also matched across matrices, with osimertinib eluting at 1.512 minutes in plasma and 1.517 minutes in cerebrospinal fluid.

Were matrix effects a problem

No. Matrix effect was calculated from six blank lots of each matrix at 3 and 800 ng/mL in plasma and 3 and 80 ng/mL in cerebrospinal fluid, and ran from 86.98 to 104.23 percent. That sits inside the accepted window for quantitative bioanalysis, so ion suppression or enhancement did not distort the reported concentrations.

How stable were the analytes during storage

Samples were held for 6 hours at 25 degrees Celsius, 30 days at minus 80 degrees Celsius, and through four freeze-thaw cycles at minus 80 degrees Celsius. Precision stayed at or below 15 percent RSD and relative deviation within plus or minus 15 percent of nominal across all three conditions, which matters most for banked retrospective samples.

Did plasma concentration predict treatment response

No. Steady-state plasma concentration showed no significant correlation with lung tumor response for either drug. Cerebrospinal fluid concentration and penetration rate did track with intracranial response, with P below 0.001 and P below 0.01 respectively, which is the central finding of the study. The response groups were small, so the finding is fragile.

What was the clearest single patient example

One patient achieved partial response in the lung while intracranial disease progressed. That patient had a cerebrospinal fluid concentration of 1.1 ng/mL and a penetration rate of 0.44 percent, so systemic exposure was adequate while central nervous system exposure was nearly absent. The same plasma sample would have looked reassuring on its own.

Does this apply to mycophenolate mofetil and valganciclovir

The principle does, even though the drugs differ. Mycophenolic acid exposure varies several-fold between patients on the same mycophenolate mofetil dose because of protein binding and enterohepatic recirculation, while valganciclovir has both an efficacy floor and a marrow toxicity ceiling. Match the matrix to the site and the window.

What are the limitations of this evidence

Twenty patients is small, with response subgroups of 3, 7 and 3 patients, and several comparisons did not reach significance. Cerebrospinal fluid is a surrogate for brain tissue rather than a direct measure, and the penetration ratio inherits the error of both measurements. No dosing rule can be drawn from association alone.

Comments are closed